Need perfume-ad references, not another random prompt?
Try commercial + close-up, then search “perfume”. Open a match to read the brief and check its source.
I maintain vflow; here's the actual lookup:
https://t.co/3eZgoZrCt0
#AIFilmmaking
For visual direction, compare source prompts by camera movement rather than “make it cinematic”.
I maintain vflow; stored writing + original-source links: https://t.co/3eZgoZrCt0
Text may omit images/settings. A reference library, not an editing service.
“Understands taste” still needs a reference brief.
For an AI-edited product ad, I’d separate:
• proof: what viewers must see
• mood: light, texture, rhythm
• constraints: logo, wording, timing
A filled example ↓
I maintain vflow; I haven’t tested this agent.
Today, we are launching VEED AI.
The world's first video editing agent that really understands taste, brand, and creative direction.
• Drop your footage or start from scratch
• Place a link to your product or service
• It trims, adds subtitles, b-rolls, motions, & SFX
Try it here: https://t.co/ysx6NGP9km
Example brief (hypothetical), coffee launch, 15s:
Proof: readable package + real pour.
Mood: warm side light, slow reveal.
Constraints: no rewritten label; last 3s for packshot.
A visual reference sets direction. Actual footage proves product behavior.
From character menu to world reveal: six beats in a shared game-trailer prompt.
Original reading diagram; timing follows the prompt, not verified output.
I maintain vflow. Read the sequence + source: https://t.co/LwkQ5VPkcq
An 'original prompt' is more than a caption.
A completed source check: shared text, missing inputs, and a 2.5/2.0 model-label conflict.
I maintain vflow. Worked reference + original source:
https://t.co/eJLKZN3mKK
#PromptEngineering
AI filmmakers: a matching face can still leave a bad cut. Which continuity problem costs you the most rework?
I maintain vflow; reference checks for your next cut: https://t.co/HC1Pyspp0A
#AIFilmmaking
@chikachiglobe I’m on the reference side of AI product ads. I maintain vflow: find examples by model/camera move, then read stored prompts and original sources. A perfume close-up is a useful exercise in separating product motion from camera motion. https://t.co/3eZgoZrCt0
Reverse time, not the train.
This prompt separates backward-running gears from the locomotive’s later forward motion, while asking for a thin, flat watch throughout.
I maintain vflow. Full prompt + source:
https://t.co/9mdi7khibT
https://t.co/6AgSFSptgy
Exact example from the walkthrough: Golden Hour Luxury Perfume Commercial.
Stored prompt + original-source link:
https://t.co/Jbw1aimLrd
Reference reading, not a generation or reproduction guarantee.
Need perfume-ad references, not another random prompt?
Try commercial + close-up, then search “perfume”. Open a match to read the brief and check its source.
I maintain vflow; here's the actual lookup:
https://t.co/3eZgoZrCt0
#AIFilmmaking
@SeaArt_Ai The 13.3–15.5s walk in the prompt gives a reset before the ridge reveal. I’d track blade position and light on the smoke across that cut. Was the wide reveal text-only, or did you also use a reference frame?
Relief at 24s. Doubt at 27s.
A horror prompt's false ending, mapped as an original diagram. Not a generated clip or model test.
I maintain vflow. Shared prompts + sources:
https://t.co/3eZgoZrCt0
Source: https://t.co/YQyE2lvEXb
30 seconds of pure Korean folk horror.
Made with Seedance 2.5 on @openart_ai
Prompt:
30-sec cinematic Korean folk-horror ritual. Keep the Word character’s face and outfit consistent throughout.
1. 0–4s: Word character kneels pale and weak in a dim traditional Korean house; worried family, candles, incense, shaman preparing the ritual.
2. 4–8s: Salt circle and talismans surround her; candlelight flickers as her eyes twitch and fingers convulse.
3. 8–12s: Her head snaps unnaturally; guttural inhuman voice, dark veins spreading, body twisting as the ritual intensifies.
4. 12–16s: Eyes turn black; shaman chants frantically and throws rice/salt; her body convulses violently.
5. 16–20s: She rises unnaturally; shaman presses a burning talisman to her forehead as wind tears through the closed room.
6. 20–24s: Talisman burns, flames surge, she collapses and becomes completely still.
7. 24–27s: Her eyes reopen human; family rushes to her in relief while the exhausted shaman watches.
8. 27–30s: Wide aftermath: scattered salt, burnt talismans, melted candles. One intact talisman emits faint dark smoke, leaving an unsettling unresolved ending.
Style: cinematic Korean folk horror, dark wooden interior, warm candlelight, deep shadows, tense atmosphere, realistic, unsettling, dramatic camera movement.
AI 영상을 연출별로 정리해둔 프롬프트 라이브러리
약 9,000개의 영상과 프롬프트가 모여있음
광고·판타지·브이로그 같은 장르부터
카메라 연출과 조명까지 필터링할 수 있어서
원하는 레퍼런스 영상을 찾기 쉬움
각 영상의 카메라 움직임, 무드 등을
시트로 정리해줘서 긴 프롬프트를 일일이 읽지 않아도
핵심 연출 요소를 한눈에 참고할 수 있음
Planning an AI music video? Give the verse and chorus a shared world: cast, wardrobe, light direction and prop ownership. Then vary the action and framing.
Original planning sketch; not a model test. I maintain vflow. More reference checks:
https://t.co/HC1Pyspp0A
@TopMediai Thanks for clarifying the joint generation. Were the orange-slice reveal and the cut points specified in the prompt, or did the model choose their timing? That would help distinguish written shot direction from output timing when studying this reference.
@tarumainfo SceneFlow’s script + playback pairing makes intent vs result inspectable. For a reference, I’d add one run row: input-image IDs | model/version | settings | script revision. It separates “borrow this shot” from “reproduce this setup.” I maintain vflow; the distinction matters.
@Enaya_ai Reading the prompt, the star’s route is continuous, while balloon reactions, bus interiors and the driver change viewpoint. For reference tagging, would you call this continuous action rather than one uninterrupted tracking shot? I maintain vflow and want to label it clearly.
@virgilerietsch Was that profile visits, site clicks, or both? What path did viewers take to the product? I want to understand which step actually worked before adapting the format for a reference library.
An AI basketball storyboard is more than a camera list.
I mapped @ShamiWeb3's prompt into action, coverage and sound. Choose the finish; don't invent shot timings.
I curate vflow. Full prompt + source:
https://t.co/d8elD1467m
#AIFilmmaking
@umesh_ai I curate vflow references. Your prompt separates eight camera-path sections within one take, two global slow-motion windows, and damage that persists. Did you block the continuous camera route before timing the action? Those are three different things to study in a chase.